A comparison of the classification accuracy of neural and neurofuzzy approaches in credit approval.

摘要

The increase in default rates and bankruptcy applications by credit customers in recent years has led financial institutions to seek better models for predicting default risks more accurately than the statistical models that have been used for many years. This has led to the use of artificial intelligence models, specifically neural networks and Neurofuzzy systems.; In this study we compare the classification accuracy of the multilayered perceptron neural network to that of the Neurofuzzy system ANFIS. The rules used for ANFIS training are derived from Classification Trees constructed using CART 5.0. The main objective is to classify credit applicants into two groups, high risk of default or low risk of default determined by relevant variables obtained from historical data.; The results obtained reveal that the neural network is a better classifier than CART based ANFIS based on the Root Mean Square Error measure. Classification accuracy measures however show that CART is superior to the neural network. We also hypothesize that it may be superior to ANFIS.